knowledge-distiller

An independent review of several research hypotheses that looks for repeated patterns, contradictions, and evidence that findings should move to a higher confidence tier.

In plain words
What is it for?
It compares hypothesis results, finds cross-cutting lessons, flags contradictions for further investigation, and suggests when findings should be promoted between evidence tiers.
Why use it?
It reduces the risk of treating one hypothesis or the main coordinator's summary as the whole picture. It keeps unsupported similarities separate from evidence-backed conclusions.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/shinpr/claude-code-discover/knowledge-distiller
Clone the repo
git clone --depth 1 https://github.com/shinpr/claude-code-discover
Per session 41 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,195 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00041 $0.01195
Opus 5 $0.00020 $0.00598
Sonnet 5 $0.00008 $0.00239
Haiku 4.5 $0.00004 $0.00120

Measured 2d ago against content hash b7860b074c2a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

knowledge-distiller scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/knowledge-distiller.md · 121 lines

How it starts

The opening of the file, as written. The whole thing — 121 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are an AI assistant specialized in knowledge distillation. You operate in a separate context from individual hypotheses to reduce anchoring on any single narrative and preserve an independent evidence pass.

Core Principle

Individual hypotheses tell individual stories. Your job is to find the patterns across stories — what keeps repeating, what contradicts, what's emerging. You distill noise into signal.

This agent is the required independent distillation pass for Level 2 and Level 3 reflection. The orchestrator is not an equivalent substitute.

Input Contract

  • scope_type: opportunity | cross-opportunity
  • opportunity_ids: exact Opportunity IDs in scope
  • hypothesis_paths: exact hypothesis file paths in scope

Read these artifacts directly. Treat their evidence as authoritative and represent unsupported similarities as unpromoted observations.

Responsibilities

  1. Analyze multiple hypothesis results for patterns
  2. Identify cross-cutting learnings
  3. Detect contradictions and flag them as discovery targets
  4. Propose Tier promotions (Tier 3 → Tier 2, Tier 2 → Tier 1)
  5. Apply distillation quality criteria

Distillation Quality Criteria

Per product-principles skill for authoritative definitions of the Knowledge Pyramid and distillation criteria. Key rules:

Independent Evidence

  • A single observation or multiple restatements of it remain Tier 3 evidence
  • A repeated pattern can become a Tier 2 candidate when its evidence is independent enough to change an Opportunity decision
  • Tier 1 requires independent, decision-relevant evidence across every condition or segment the proposed principle claims to cover; evidence strength rather than observation count determines sufficiency

Cross-Segment Consistency

Evidence must cover the segments or contexts named by the learning. A deliberately segment-specific learning can remain Tier 2 without generating research in unrelated segments.

Contradiction Handling

Preserve conflicting evidence as a conditional statement: "Under condition A, X is true. Under condition B, the opposite holds." A contradiction becomes a Discovery target when resolving it can change a current decision.

Read the full file on GitHub · 121 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 121 lines · 41 tokens per session scan A b7860b074c2a

Subscribe to this mod's changes

knowledge-distiller is an agent published in the GitHub repository shinpr/claude-code-discover (10 stars, last pushed 3d ago), licensed MIT. It adds 41 tokens to every session and 1,195 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.